Renesas announces memory technology for AI
Renesas Electronics has developed an AI accelerator that performs convolutional neural network (CNN) processing at high speeds and low power. A test chip with this accelerator has achieved the power efficiency of 8.8Tera operations per second per W (TOPS/W), which is the industry’s highest class of power efficiency, reports Renesas. The accelerator is based on the processing-in-memory (PIM) architecture, in which multiply-and-accumulate (MAC) operations are performed in the memory circuit as data is read out from that memory.
To create the new AI accelerator, Renesas developed three technologies. The first is a ternary-valued (-1, 0, 1) SRAM structure PIM technology that can perform large-scale CNN computations. The second is an SRAM circuit to be applied with comparators that can read out memory data at low power. The third is a technology that prevents calculation errors due to process variations in the manufacturing. Together, these technologies achieve a reduction in the memory access time in deep learning processing and a reduction in the power required for the MAC operations. As a result, the accelerator achieves the industry’s highest class of power efficiency while maintaining an accuracy ratio more than 99 per cent when evaluated in a handwritten character recognition test (MNIST), claims Renesas.
Before this development, the PIM architecture was unable to achieve an adequate accuracy level for large-scale CNN computations with single-bit calculations because the binary (0,1) SRAM structure was only able to handle data with values 0 or 1. Additionally, process variations in the manufacturing reduced the reliability of these calculations. The technologies developed by Renesas resolve these issues and can be applied to implement AI chips of the future and e-AI solutions for applications such as wearable equipment and robots that require both performance and power efficiency, says Renesas.
Since introducing the embedded AI (e-AI) concept in 2015, Renesas has defined classes based on the effectiveness of e-AI and applications that are implemented and has been developing e-AI solutions based on four classes: judging the correctness or abnormality of signal waveform data; judging correctness or abnormality using real-time image processing; performing recognition in real time and enabling incremental learning at an endpoint.
